[Libreoffice-bugs] [Bug 130725] ImpSvNumberInputScan::StringToDouble may produce inaccurate result

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=130725

--- Comment #32 from Mike Kaganski <mikekaganski at hotmail.com> ---
(In reply to b. from comment #31)

1. I myself marked my response as off-topic, to not clutter the issue with
futher unrelated comments. Of course, that doesn't stop you from keeping doing
just that here, as you do in multiple other places. OK, I will repeat the
labeling again. I reserve the right to keep hoping that someday you might
finally realize that your random spam in random places is not welcome.

2. No matter if you are programmer or not. You are wasting everyone's time
doing what you do. It's *your* duty to provide at least a proof of correctness
of your ideas, if they don't immediately meet support in developers. So no, I'm
not going to invent and run systematic tests to see if some random stuff you
suggest has flaws or actually improves things.

You are using logic fallacies right and left, klike declaring a necessity of "a
change of thinking ... to 'ok, there are some difficulties, what is the best
way to deal with them'", which is falsely declaring your opponents not having a
positive attitude, despite them (me and Eike specifically) having implemented
multiple improvement just in that area, and Eike pointed to other things like
*actually proven correct* Kahan algorithm. But your phrases may create a
perception in a random reader that you suggest some brilliant revolutionary
concept, but fall victim of inert developers. Nice way of argument, bravo!

If you believe your "ideas or recipes" have any value, then just do the
thorough tests. And that is not "I take 5 or 50 chosen problematic cases, and
see them improved", but "I get 4 billion floats (32-bit floating IEEE 754 point
numbers), and simulate the calculations according to my formula on them, and
compare to the raw calculations like Calc does; then use 64-bit doubles to
perform the same calculations, and compare the error from the two calculations
above, to prove that my idea actually has better result over the whole range of
tested numbers. That took some days/weeks of programming, and hours or
runtime". That would be a reasonable effort, showing some value in checking
your testing procedure, and then checking the viability of the result. Not "I
give you something that is obvious to you as useless and wrong, and repeatedly
expect you to spend your time think over it, and either test, or describe why
that is wrong, which I will happliy ignore again."

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20201217/ef99515b/attachment-0001.htm>